Dual Path Transformer

Dual-path transformers are a class of neural network architectures designed to leverage both local and global information within data, improving performance on various tasks compared to single-path approaches. Current research focuses on applying this architecture to diverse domains, including speech enhancement, image generation, and point cloud processing, often incorporating techniques like multi-head attention and convolutional layers for enhanced feature extraction and processing. These models demonstrate improved accuracy and efficiency in tasks ranging from audio signal processing to computer vision, highlighting the versatility and potential of dual-path transformers for complex data analysis. The resulting advancements have significant implications for various fields, including medical imaging, autonomous driving, and natural language processing.

Papers

September 2, 2024

Spectron: Target Speaker Extraction using Conditional Transformer with Adversarial Refinement
Tathagata Bandyopadhyay
Speech Encoder Refinement Network Target Speaker Extraction Attention Based Transformer Dual Path Transformer Conditional Transformer

October 30, 2023

DCHT: Deep Complex Hybrid Transformer for Speech Enhancement
Jialu Li, Junhui Li, Pu Wang, Youshan Zhang
Speech Enhancement Cross Domain Mel Spectrogram Hybrid Transformer Dual Path Transformer

August 30, 2023

Dual-path Transformer Based Neural Beamformer for Target Speech Extraction
Aoqi Guo, Sichong Qian, Baoxiang Li, Dazhi Gao
Adaptive Beamforming Target Speech Extraction Dual Path Transformer Neural Beamformer

July 17, 2023

SVDFormer: Complementing Point Cloud via Self-view Augmentation and Self-structure Dual-generator
Zhe Zhu, Honghua Chen, Xing He, Weiming Wang, Jing Qin, Mingqiang Wei
Point Cloud Multi View Point Cloud Completion Self Augmentation Incomplete Point Cloud Dual Path Transformer Partial Shape

May 24, 2023

Dual Path Transformer with Partition Attention
Zhengkai Jiang, Liang Liu, Jiangning Zhang, Yabiao Wang, Mingang Chen, Chengjie Wang
Attention Block Dual Path Transformer

May 16, 2023

DualGenerator: Information Interaction-based Generative Network for Point Cloud Completion
Pengcheng Shi, Haozhe Cheng, Xu Han, Yiyang Zhou, Jihua Zhu
Full Information Point Cloud Completion Generative Network Incomplete Point Cloud Dual Path Transformer Path Dependent

May 10, 2023

Multi-Path Transformer is Better: A Case Study on Neural Machine Translation
Ye Lin, Shuhan Zhou, Yanyang Li, Anxiang Ma, Tong Xiao, Jingbo Zhu
Case Study Transformer Model Neural Machine Translation Deeper Model Multi Path Large Scale Transformer Dual Path Transformer

April 11, 2023

OccFormer: Dual-path Transformer for Vision-based 3D Semantic Occupancy Prediction
Yunpeng Zhang, Zheng Zhu, Dalong Du
Semantic Segmentation Semantic Scene Completion Semantic Occupancy 3D Semantic Occupancy Dual Path Transformer

March 2, 2023

Meta-information-aware Dual-path Transformer for Differential Diagnosis of Multi-type Pancreatic Lesions in Multi-phase CT
Bo Zhou, Yingda Xia, Jiawen Yao, Le Lu, Jingren Zhou, Chi Liu, James S. Duncan, Ling Zhang
Differential Diagnosis Multi Phase Dual Path Transformer Pancreatic Ductal Adenocarcinoma Pancreatic Tumor

February 23, 2023

MossFormer: Pushing the Performance Limit of Monaural Speech Separation using Gated Single-Head Transformer with Convolution-Augmented Joint Self-Attentions
Shengkui Zhao, Bin Ma
Self Attention Head Transformer Attention Based Sequence Dual Path Transformer Self Attention Computation Monaural Speech Separation Performance Limit

November 18, 2022

DGD-cGAN: A Dual Generator for Image Dewatering and Restoration
Salma Gonzalez-Sabbagh, Antonio Robles-Kelly, Shang Gao
Underwater Image Conditional Generative Adversarial Network Image Deraining Restoration Quality Underwater Image Formation Dual Path Transformer

October 30, 2022

TT-Net: Dual-path transformer based sound field translation in the spherical harmonic domain
Yiwen Wang, Zijian Lan, Xihong Wu, Tianshu Qu
Sound Field Spherical Harmonic Dual Path Transformer Spherical Harmonic Domain Radial Function

August 21, 2022

DPTNet: A Dual-Path Transformer Architecture for Scene Text Detection
Jingyu Lin, Jie Jiang, Yan Yan, Chunchao Guo, Hongfa Wang, Wei Liu, Hanzi Wang
Deep Convolutional Neural Network Scene Text Detection Multi Path Dual Path Transformer

August 6, 2022

SSDPT: Self-Supervised Dual-Path Transformer for Anomalous Sound Detection in Machine Condition Monitoring
Jisheng Bai, Jianfeng Chen, Mou Wang, Muhammad Saad Ayub, Qingli Yan
Acoustic Feature Anomalous Sound Detection Acoustic Representation Machine Condition Monitoring Dual Path Transformer Anomalous Sound

February 17, 2022

3D-Aware Indoor Scene Synthesis with Depth Priors
Zifan Shi, Yujun Shen, Jiapeng Zhu, Dit-Yan Yeung, Qifeng Chen
Generative Adversarial Network Image Synthesis Dual Path Transformer Indoor Scene Synthesis

December 21, 2021

MPViT: Multi-Path Vision Transformer for Dense Prediction
Youngwan Lee, Jonghee Kim, Jeff Willette, Sung Ju Hwang
Vision Transformer Dense Prediction Multi Scale Representation Dual Path Transformer Multi Scale Feature Representation

Dual Path Transformer

Papers

Spectron: Target Speaker Extraction using Conditional Transformer with Adversarial Refinement

DCHT: Deep Complex Hybrid Transformer for Speech Enhancement

Dual-path Transformer Based Neural Beamformer for Target Speech Extraction

SVDFormer: Complementing Point Cloud via Self-view Augmentation and Self-structure Dual-generator

Dual Path Transformer with Partition Attention

DualGenerator: Information Interaction-based Generative Network for Point Cloud Completion

Multi-Path Transformer is Better: A Case Study on Neural Machine Translation

OccFormer: Dual-path Transformer for Vision-based 3D Semantic Occupancy Prediction

Meta-information-aware Dual-path Transformer for Differential Diagnosis of Multi-type Pancreatic Lesions in Multi-phase CT

MossFormer: Pushing the Performance Limit of Monaural Speech Separation using Gated Single-Head Transformer with Convolution-Augmented Joint Self-Attentions

DGD-cGAN: A Dual Generator for Image Dewatering and Restoration

TT-Net: Dual-path transformer based sound field translation in the spherical harmonic domain

DPTNet: A Dual-Path Transformer Architecture for Scene Text Detection

SSDPT: Self-Supervised Dual-Path Transformer for Anomalous Sound Detection in Machine Condition Monitoring

3D-Aware Indoor Scene Synthesis with Depth Priors

MPViT: Multi-Path Vision Transformer for Dense Prediction